Figure 4

DCS led to spontaneous excitatory and inhibitory effects in corticostriatal neurons and DCS-based learning resulted in changes in neuronal ensemble activity. (a) Representative separation of single unit cluster (yellow) from stimulation artifacts (green) projected on the first two principle components during offline sorting. (b) Single unit waveforms were clearly distinguishable from the stimulation artifacts. (c) Stimulation waveforms resulted in amplifier dead time due to saturation during the ‘stimulation on’ period. A conservative blanking period of 4 msec was heuristically calculated by observing the time window between artifact and spike (bottom). (d) Spikes from same neuron after artifact removal (top). Peri stimulus time histogram (PSTH) of neuron before and after correction of firing rate with artifact blanking method (bottom). (e) Representative PSTH of M1, S1 and Str neurons showing excitatory (top row) and inhibitory (bottom row) responses during SCS delivery (brown bar) as compared to baseline period (green bar). Red and blue dashes over PSTHs show significant response periods. Magenta dotted line indicates mean firing rate during baseline. (f) Mean Z-scored firing rates of M1, S1, Str neurons with excitatory and inhibitory responses in early and late sessions (n = 7 rats). (g) Z-scored firing rates of all neurons recorded during early and late sessions sorted by onset of excitatory (red star) and inhibitory (blue star) response. Vertical bars indicate neurons separated by response type - red: excitatory, blue: inhibitory, magenta – mixed, black: no response. (h) Peri-stimulus z-scored firing rates of M1, S1 and STR for a representative rat during early and late training sessions. (Red bar: M1 neurons, violet bar: S1 neurons, and magenta bar: STR neurons. (i) Mean firing rate for the neuronal ensemble in ‘h’ showed clear differences between early (green) and late (blue) sessions. (j,k) To quantify changes in ensemble activity between early and late sessions, neuronal PSTHs were fed to a custom logistic regression classifier for predicting the stimulus type (‘virtual narrow’ vs ‘virtual wide’). Adaptive decoding i.e. retraining the classifier on each trial by updating the training set before making a prediction resulted in higher accuracy and lower mean error than non-adaptive decoding (n = 6 rats). (l,m) There was significant improvement in the decoding accuracy and mean error in late sessions as compared to early sessions (n = 6 rats). For all panels, time (0) corresponds to stimulation onset, brown horizontal bar indicates stimulation period, shaded areas and error bars are s.e.m., (*p < 0.05).